Segmentation of Normal and Pathological Tissues in MRI Brain Images Using Dual Classifier

S.Javeed Hussain 1, C.Venkatesh 2+, S. Asif hussain 2, L.Chetana 2 and V.Gireesha 2 1 BCETFW, Kadapa, A.P., India

2 AITS, Rajampet, A.P., India.

Abstract. In this paper, an efficient technique is proposed for the precise segmentation of normal and pathological tissues in the MRI brain images. The proposed segmentation technique initially performs classification process by utilizing FFBNN. Dual FFBNN networks are used in the classification process. The inputs for these networks are the features that are extracted in two ways from the MRI brain images. Five features are extracted from the MRI images: they are two dynamic statistical features and three 2D wavelet decomposition features. In Segmentation, the normal tissues such as WM (White Matter), GM (Gray Matter) and CSF (Cerebrospinal Fluid) are segmented from the normal MRI images and pathological tissues such as Edema and Tumor are segmented from the abnormal images. The non-cortical tissues in the normal images are removed by the preprocessing stage. The performance of the segmentation technique is evaluated by performance measures such as accuracy, specificity and sensitivity. The performance of segmentation process is analyzed using a defined set of MRI brain.

Keywords: Edema, Cortical tissues, dynamic, MRI, Segmentation

1. Introduction Segmentation of brain tissue on magnetic resonance (MRI) images normally determines the type of

tissue present for each pixel or voxel in a 2D or 3D data set respectively, based on the information gathered from both MR images and prior knowledge of the brain. Segmentation at preliminary stage is important and necessary for the analysis of medical images for computer-aided diagnosis and treatment. Magnetic resonance imaging (MRI) is a significant diagnostic imaging method for non-invasive imaging.

The brain matters are mainly categorized as white matter, gray matter, cerebrospinal fluid (CSF) or vasculature. Magnetic resonance imaging (MRI) systems can generate many images of inner anatomical structures in the same body section with multiple differences, based on the local variations of spin–spin relaxation time (T2), spin–lattice relaxation time (T1), and proton density (PD) [5]. The presence of noise, errors in the scanners, and the structural variations of the imaging objects are the major obstruction to the segmentation of 1MR images.

2. Proposed Methodology for Tissue Segmentation in MRI Brain Images In this paper, we propose an efficient method to segment the normal and pathological tissues in the MRI

brain images. Two major stages are involved in our proposed methodology: o Classification o Segmentation

+ Corresponding author. Tel.: + 919985032919. E-mail address: venky.cc@gmail.com.

165

2011 International Conference on Advancements in Information Technology

With workshop of ICBMG 2011

IPCSIT vol.20 (2011) © (2011) IACSIT Press, Singapore

2.1. Classification In classification stage, the MRI brain images are classified into normal and abnormal brain images.

Two phases are involved in this classification that are mentioned below (i) Feature extraction (ii) Network training and testing

2.2. Segmentation Segmentation process is performed in both normal and abnormal images. In normal images, the normal

tissues such as WM, GM and CSF are segmented and in abnormal images, the edema and tumor tissues are segmented. Following are the two steps involved in the segmentation process: i) Preprocessing ii) Tissue Segmentation iii) Normal tissue segmentation iv) Pathological tissue Segmentation 2.2.1． Preprocessing

Among all preprocessing methods, Skull stripping is used for the segmentation of brain tissues. In skull stripping, initially the given MRI brain image is converted into gray scale image and then a morphological operation [4] is performed in the gray scale image. Then the brain cortex in the gray scale image is stripped by using region based binary mask extraction. 2.2.2． Tissue Segmentation

After skull stripping, the brain MRI images are involved in the tissue segmentation of segment the WM, GM, CSF, and edema 2.2.3. Normal Tissue Segmentation

Segmentation of Normal tissues such as WM, GM and CSF are performed from the normal images. Here, segmentation process is performed in two ways namely, (i) WM and GM segmentation (ii) CSF segmentation

• WM and GM segmentation The skull stripped image sI is given as input to the WM and GM segmentation process. The gradient of

two variables x and y is defined as follows, ∧∧

∂∂

+∂

∂=∇ j

yI

ix

IyxI GGG ),(

(1)

Using the gradient values, the current edges in the image are marked using the Equ. (2) & (3). 2)(2)( ji yxG += (2)

G

E m +=

11 (3)

Then, the binarization process is performed in the edge marked image mE Opening and closing operation is

utilized ⎪⎩

⎪⎨⎧

=

==

0;

1;

i

i

b

bwg IifGM

IifWMI (4)

• CSF Segmentation To segment the cerebrospinal fluid from the brain MRI image, an Orthogonal Polynomial Transform

(OPT) is applied to the skull stripped image sI is computed using the following formula,

( )23

( ) 0 . 0 5 * ( | | )1 0 0

s ic f s

II S i n r a n d I

⎛ ⎞= +⎜ ⎟⎜ ⎟

⎝ ⎠

(5)

2.2.4. Pathological Tissue Segmentation Pathological tissues such as edema and tumor are segmented from the classified abnormal images and

these tissues are segmented by two different methods: (i) Tumor ii) Edema • Tumor Segmentation

The tumor tissue segmentation is performed in the abnormal brain MRI images. The RGM observes the neighbor pixel values with the initial seed points, that is it checks tumor segmentation result is represented as TI .

• Edema Segmentation Edema tissue is segmented from the abnormal image aI . Each pixel in the image is compared with these

threshold values to select the pixels

166

⎩⎨⎧ ≥≤

=otherwise

tttppX uu

;0&,; 453 (6)

Experimental Results The proposed brain tissue segmentation technique is implemented in the working platform MATLAB

(version 7.10) and it is evaluated using 10 medical brain MRI images, which are collected from various medical diagnosis centers. Among 10 MRI images, 5 images are normal and the remaining is abnormal.

Fig. 1: The sample input of normal and abnormal images

The input images are classified by two FFBNN networks. Input values for both FFBNN networks are five features such as mean, variance, horizontal, vertical and diagonal functions of 2D wavelet decomposition and these features is given as input to the dual FFBNN networks. The classification results of dual FFBNN networks are shown in Fig.1. Then, the segmentation process is performed on the classified images. The normal images are segmented into three normal tissues such as WM, GM and CSF and the abnormal images are segmented into two pathological tissues such as edema, tumor. The segmented normal tissue results are shown in Fig. 2. The intermediary result of the edema segmentation is shown in Figure 3. The segmented pathological tissues are shown in Figure 4.

Fig. 2: Outputs of normal tissues (i) WM (ii) GM (iii) CSF (iv) WM, GM and CSF

167

Fig.3: (i) Histogram Equalized image (ii) HSVmodel (iii)HSVThresholding (IV) closing

(v) Edema region (VI) Closing (vii) Dilation

Fig. 4: Segmentation result of pathological tissues

(i) Tumor (ii) Edema and (iii) Tumor and Edema in abnormal image

3. Conclusion In this paper, an efficient segmentation was developed to segment the normal and pathological tissues

from the MRI brain images. The performance of the proposed segmentation was analyzed using defined set of MRI normal and abnormal images. The performance of the method was understood from the experimental results and analysis. The proposed tissues segmentation method performance is evaluated with the aid of five images. The normal WM, GM and CSF tissues segmentation is of 99%, 82% and 99% mean accuracy results respectively. The higher accuracy performance gives more precise segmentation results in the normal images. Furthermore, pathological tissues edema and tumor also gives 98%, 93% mean accuracy results respectively. Hence the performance of our proposed tissues segmentation method gives more efficient and effective results in both normal and pathological tissues segmentation process.

4. Acknowledgements We would like to thank Y.Sirian R&D from CHENNAI, for their valuable suggestions given in

implementing the project and ECE Dept., AITS, Rajampet for their overall help and guidance.

5. References [1] Chaozhe Zhu and Tianzi Jiang, "Multicontext Fuzzy Clustering for Separation of Brain Tissues in Magnetic

Resonance Images", NeuroImage, Vol.18, No. 3, pp. 685-696, 2003

[2] Shan Shen, William Sandham, Malcolm Granat and Annette Sterr, "MRI Fuzzy Segmentation of Brain Tissue Using neighborhood Attraction With Neural-Network Optimization", IEEE Transactions On Information Technology In biomedicine, Vol. 9, No. 3, pp. 459-467, September 2005

[3] Senthilkumaran and Rajesh, "Brain Image Segmentation using Granular Rough Sets", International Journal of

168

Arts and Sciences, Vol. 3, No. 1, pp. 69 - 78, 2009

[4] Pradipta Maji, Malay K. Kundu and Bhabatosh Chanda, "Second Order Fuzzy Measure and Weighted Co-occurrence Matrix for Segmentation of Brain MR Images", Journal of Fundamenta Informaticae, Vol. 88, No. 1-2, pp. 161-176, 2008

[5] Jzau-Sheng Lin, Kuo-Sheng Cheng, and Chi-Wu Mao, "Segmentation of Multispectral Magnetic Resonance Image using Penalized Fuzzy Competitive Learning Network", Journal of Computers and Biomedical Research, Vol. 29, No. 4, pp. 314–326, 1996

[6] Mostafa G. Mostafa, Mohammed F. Tolba, Tarek F. Gharib and Mohammed A-Megeed, "A Gaussian multiresolution Algorithm For Medical Image Segmentation", In Proceedings of IEEE International Conference On intelligent Engineering Systems, Assiut-Luxor, Egypt, 2003

[7] Jagath C. Rajapakse, Jay N. Giedd and Judith L. Rapoport, "Statistical Approach to Segmentation of Single-channel cerebral MR Images", IEEE Transactions on Medical Imaging, Vol. 16, No. 2, pp. 176-186, April 1997

169